The Cloud Population and Onset of the Madden-Julian Oscillation over the Indian Ocean during dynamo-amie

The Cloud Population and Onset of the Madden-Julian Oscillation over the Indian Ocean during DYNAMO-AMIE

Scott W. Powell and Robert A. Houze, Jr.

Submitted to the Journal of Geophysical Research - Atmospheres

Revised September 2013


Corresponding Author: Scott W. Powell, Department of Atmospheric Sciences, University of Washington, Box 351640, Seattle, WA 98195

Email: spowell@atmos.uw.edu

Key Points

• 
  MJO onset characterized by sharp increase in stratiform precipitation .
  
• 
  Humidity and convective depth also increase over 3-7 days at MJO onset.
  
• 
  Convection contributes to tropospheric moistening prior to MJO onset.
  

Variability of the cloud population in the central-equatorial Indian Ocean was observed in the context of the Madden-Julian Oscillation (MJO) during the Dynamics of the Madden-Julian Oscillation (DYNAMO) and Atmospheric Radiation Measurement (ARM) Madden-Julian Investigation Experiment (AMIE) field campaigns. Radar observations from the S-PolKa radar system on Addu Atoll in the Maldives characterize the types of convective and stratiform radar echoes and the heights they reach. To gain insight into the relationship between clouds and humidification of the troposphere leading up to and during an active MJO event, the work compares relates variability of the observed precipitation structure to that of tropospheric humidity and upper-level zonal wind using highly resolved and co-located rawinsonde measurements. The variability in stratiform precipitation areas dominates variability in the nature of precipitating convectionvariation in convective behavior associated with the MJO. Areal coverage of precipitating radar echo, convective echo-top height, and tropospheric humidity above 850 hPa rapidly increase over ~3-7 days near MJO onset. This rate of increase is substantially faster than the 10-20 days needed for build-up of moisture prior to MJO onset as hypothesized by the "discharge-recharge" hypothesis. Convective echoes become more common during the days prior to MJO onset, and the increased convection occurs before low-tropospheric moistening. The upper troposphere rapidly moistens as the first widespread stratiform region passes over an area. Thus, clouds likely play a role in tropospheric humidification. Whether increased low-tropospheric humidity causes vertical growth of convection ishas not yet determined.

AGU: 3314 Convective Processes, 3329 Mesoscale Meteorology, 3371 Tropical Convection, 3373 Tropical Dynamics, 9340 Indian Ocean

Author key words: Madden-Julian Oscillation, Tropical Convection, DYNAMO, discharge-recharge, AMIE, convection over Indian Ocean

1. Introduction

Since the late 1960s, field projects deployed in low latitudes have yielded information about the structure and behavior of deep convection, and in the tropics geostationary satellites have provided a global view via visible and infrared sensing. During the past 15 years, passive microwave and active sensors (radar and lidar) aboard Earth-orbiting platforms have observed tropical clouds and precipitation in greater detail. Despite these years of intensive observations of tropical convection, one dominant mode of tropical convection occurring on intraseasonal timescales (30 to 90 days) remains poorly understood.

Hypotheses explaining the mechanisms through which the onset of convection occurs over the Indian Ocean include Knutson and Weickmann's [1987] idea that circumnavigating upper-tropospheric velocity potential and zonal wind anomalies occurring as a Kelvin wave response to a previous MJO may trigger the next MJO. Seo and Kim [2003] conclude that the Kelvin wave response to one MJO circumnavigates the globe and generates a new cycle. Virts and Wallace [2010] support this idea by showing evidence that a circumnavigating Kelvin wave associated with the MJO exists and modulates the frequency of cirrus in the tropical tropopause transition layer (TTL). However, while Matthews [2000] suggests that mean sea-level pressure anomalies associated with a Kelvin mode circumnavigate and coincide with the beginning of the next MJO cycle, Matthews [2008] conducts empirical orthogonal function (EOF) analysis on filtered OLR anomalies and shows that for only 60% of MJO cases can the negative anomalies be definitively traced back to a prior MJO event. More recently, a modeling study by Ray and Li [2013] suggests that, in the long-term mean sense, extratropical influences have a greater influence on generating an MJO than tropical influences such as circumnavigating Kelvin modes. StillNonetheless, such circumnavigating features are responsible for at least some individual cases of MJO onset.

One hypothesis regarding how the recharge moistening occurs is that cumulus clouds—and particularly cumulus congestus clouds [Johnson et al., 1999]—are responsible for "pre-conditioning" or "recharging" the troposphere prior to an MJO over timescales of several weeks. Benedict and Randall [2007] provide moreadd detail on to the proposed interaction between cumulus clouds and tropospheric moistening by on the basis of compositing satellite and reanalysis data relative to a filtered maximum in rainfall during several MJO events. Their results appear to support the "discharge-recharge" process. They suggest that low-level heating and moistening by cumulus clouds gradually condition the lower troposphere for explosive convective development over a 10 to 15 day periods prior to MJO onset. A positive feedback process is suggested in which allows the clouds to grow taller as moist static energy (MSE) is transported vertically; the deeper clouds then transport moist static energy to even higher levels, and the process continues until the environment is moist enough through a deep layer for widespread convective events to occur. Kemball-Cook and Weare [2001] illustrate suggest a similar process occurring over the 20 days prior to MJO onset. Support for this hypothesis has been shown offered in numerous other composite studies of sounding, satellite, model, or reanalysis data [e.g. Kiladis et al., 2005; Masunaga et al., 2006; Haertel et al., 2008; Wu and Deng, 2013].

Recent studies of tropical cumulus suggest that on time scales of hours to two or three days, pre-moistening of the free troposphere by shallow convection and congestus cannot explain rapid transitions to deep convection. Hohenegger and Stevens [2013] find that over the tropical Atlantic, local cumulus congestus clouds transition into deep convection occurs more quickly than the time needed for congestus clouds to moisten the troposphere. Furthermore, they show that the probability of a congestus cloud developing into a cumulonimbus does not increase for longer-lived congestus. Masunaga [in press] shows that for congestus clouds and mesoscale convective systems throughout the tropics, free-tropospheric moistening occurs as early as a day before convective clouds begin to deepen—or 2-3 days before the peak of cloud cover associated with a mesoscale system. While both show that humidification of the environment is not necessary for deep cloud development locally, neither study excludes the possibility that pre-moistening could promote deep and widespread convection on longer timescales, such as that proposed to be relevant to the MJO.

An alternative hypothesis is that large-scale moistening may occur gradually prior to MJO onset at one a given location as a result of horizontal moisture advection, particularly in the lower troposphere. The increase of lower tropospheric moisture would favor development of shallow convection, which would act as agents to transport MSE vertically. The largest term in the large-scale composite MSE budget over the Indian Ocean on an intraseasonal time scale is horizontal advection [Maloney, 2009]. Andersen and Kuang (2012) examine the effects of the advection of MSE on MJO propagation and again show that horizontal advection dominates the MSE budget. MSE build-up occurs to the east of convection, and drying occurs to the west of, or behind, the convection. The advection of MSE may be related to boundary layer moisture convergence, which has been hypothesized to influence the "discharge-recharge" cycle as well [Hendon and Salby 1994; Maloney and Hartmann 1998; Matthews 2000; Seo and Kim 2003]. Such studies provide a potential mechanism for why new convection is favored downstream of currently active convection. In other words, they potentially explain the eastward propagation of the MJO. They do not, however, explain how MJO onset suddenly occurs over the Indian Ocean in the absence of upstream convection.

Instances of the MJO were observed during the Tropical Ocean Global Atmosphere Coupled Ocean-Atmosphere Response Experiment [TOGA COARE; Godfrey et al., 1998] in 1992-1993 and the Mirai Indian Ocean Cruise for the Study of the MJO-convection Onset [MISMO; Yoneyama et al., 2008] in 2006. TOGA COARE provided detailed observations in segments of two MJO events over the tropical west Pacific, and MISMO documented one weak MJO event over the Indian Ocean. In TOGA COARE, rawinsondes showed humidification of the troposphere [Lin and Johnson, 1996]. Katsumata et al., [2009] attributed humidification of the upper-troposphere in MISMO primarily to eastward propagating mesoscale cloud systems.

In 2011-2012, a major field program obtained observations that are an important follow-on to TOGA COARE and MISMO. The Dynamics of the Madden-Julian Oscillation (DYNAMO) and the ARM (Atmospheric Radiation Measurement) Madden-Julian Oscillation Experiment (AMIE) were the U.S. contributions to an international collaboration studying the cloud population of the Indian Ocean in the context of the MJO [Yoneyama et al. 2013]. This cooperative experiment was unprecedented because of the amount and duration of intensive radar and rawinsonde observations over and around the Indian Ocean during boreal winter. The long period of intensive observations allowed for detailed observations of the variation of convection throughout several cycles of the MJO in an area where the initial onset of MJO-related convection frequently occurs. One major goal of the experiment was to explore hypothesized mechanisms for onset and propagation of an MJO in and through the Indian Ocean and tropical west Pacific. To date, such mechanisms are poorly understood and not well-represented in global models.

This paper examines radar and rawinsonde datasets collected during DYNAMO/AMIE in the central equatorial Indian Ocean to:

1. 
   Describe the structure and organization of convection prior to, during, and after widespread convective events associated with the MJO. Describe the structure, development, and organization of MJO-related convection over the central equatorial Indian Ocean.
   
2. 
   Investigate the relationship between vertical growth of convective clouds, humidification of the troposphere, and MJO onset where MJO-related convection initiates..
   

The intensive observing period (IOP) for DYNAMO/AMIE spanned the period 1 October 2011 through 9 February 2012. Several scanning precipitation and vertically pointing cloud radars were deployed at locations around the central equatorial Indian Ocean during DYNAMO/AMIE. Observations by these radars provide an unprecedented view of the three-dimensional structure of the convective entities making up the population of clouds in the MJO. Figure 1 shows the locations of each radar platform and rawinsonde launch site in the IOP area. At Addu, three radar sites were located within 13 km of each other: the C-band SMART-R; the ARM mobile facility (AMF2), which housed a vertically pointing cloud radar; and the dual-wavelength Doppler polarimetric S-PolKa radar system described below. The extensive radar data are supplemented by soundings taken every 3 h at the primary DYNAMO/AMIE sites and at lower frequencies at other locations throughout the Indian Ocean and maritime and Indian subcontinents. The three-hourly rawinsonde launches at Gan Island (0.69°S, 73.15°E) on Addu Atoll provided details on the vertical profiles of wind, temperature, and moisture content in the immediate vicinity of the S-PolKa radar. Detailed information about the sondes, including bias correcting and smoothing of the sonde data, is available in Long and Holdridge [2012].

For this study, we concentrate on the radar and sounding observations made on Addu Atoll. The primary radar system used for our analysis is the dual-wavelength, dual-polarimetric S-PolKa from the National Center for Atmospheric Research (NCAR). The system consists of two radars, awith wavelengths of 10 cm wavelength (S-band) and a 0.8 cm wavelength (K_a-band), radar mounted on the same platform with matching beam widths of 0.91 degrees. S-PolKa employed executed two types ofa scan strategy ies during its campaign. The radarconsisting of executed surveillance scans at elevation angles up to 11 degrees for 5 min. For, followed by the next 10 min of, the radar performed range height indicator scans (RHIs, i.e. scanning in elevation at a fixed azimuth) spaced at intervals of 2° of azimuth in the northeast and southeast sectors of radar coverage. Each RHI sector was located over open ocean, and scanned as high asup to 41°. The scanned sectors were bounded by azimuths 6 and 82°, and 114 and 141.9° respectively. The final RHI at 141.9° was performed over AMF2. Then the cycle ofof the scan sequence was scanning repeated continuously from 28 September 2011 through 15 January 2012, uninterrupted with operator-selected scans. Technical issues caused only a few minor interruptions. The RHIs were performed to document the structure of the echoes with as much vertical resolution as possible. The radar repeated nearly without exception from 28 September 2011 until 15 January 2012 to guaranteeThus, the radar obtained a statistically homogeneous dataset for all time periods, whether suppressed, active, or intermediate in terms of the cloud population, defined herein as the entire ensemble of clouds observable by S-PolKa. More information on the S-PolKa system, as well as the available datasets may be found at http://www.eol.ucar.edu/projects/dynamo/spol/SpolKa_DYNAMO_UsersGuide.toc.html.

3. Radar-derived products

Before generating allderiving products from the S-PolKa data, the reflectivity field was interpolated to a Cartesian grid with 500 m horizontal and vertical resolution using both surveillance scans and RHIs. The areal coverage of the gridded dataset is 70685 km².

Radar reflectivity returns values measured by S-PolKa were analyzed to determine if a volume of radar echo was from the S-band radar that indicate a precipitating cloud are classified as either convective or stratiform. Convective precipitation areas are characterized by strong vertical motion, heavy rain, and a low- to mid-level maximum in diabatic heating, depending on cloud depth. Stratiform regions are characterized byhave less intense vertical motion, lighter precipitation, a radar bright band near the 0˚C level where melting precipitation occurs, and a diabatic heating maximum (minimum) in the upper (lower) troposphere. The bright band is only a few hundred meters thick, and therefore the radar only resolves it well near the radar, before the beam widens too much with distance from the antenna. The algorithm used to classify clouds observed as either convective or stratiform therefore emphasize the horizontal variation of radar echo intensity. The algorithm used here is a modification of the method used by Houze [1973], Churchill and Houze [1984], Steiner et al. [1995], and Yuter and Houze [1997]. The algorithm, including input parameters used, is described briefly by Houze [1997] and in greater detail in Appendix A of Didlake and Houze [2009]. For our analysis, we use radar reflectivity at 2.5 km. Background reflectivity was computed using the mean reflectivity of all pixels within 11 km of each data point. Echoes at 2.5 km exceeding 40 dBZ are automatically classified as convective, and echoes weaker than 5 dBZ are not classified. These choices have been calibrated by manually, as advocated by Steiner et al. [1995], by checking algorithm output against RHI observations close to the radar, where it is possible to determine whether or not the algorithm is misclassifying echoes with brightbands as convective. The 11 km radius and 40 dBZ threshold values minimize such misclassification.

The radar-derived precipitation rates computed by S-PolKa utilize reflectivity (Z), differential reflectivity (Z_DR), and the specific differential phase (K_DP) of the S-band. A "hybrid" rain rate algorithm is employed to determine the best relationship among Z, Z_DR, K_DP, and rain rate to use. A detailed explanation of the algorithm with references included may be found at http://www.eol.ucar.edu/projects/dynamo/spol/parameters/rain_rate/rain_rates.html.

Echo-top heights are computed only for high-elevation RHI scans because they capture the top of all echoes except those immediately nextclose to the radar. S-PolKa surveillance scans, which were confined to lower elevation angles, as noted above, did not detect always detect the tops of echoes during many of the surveillance scans, which were confined to lower elevation angles, as noted above. In this paper we determine the maximum heights of each contiguous echo surrounded by the The products give the maximum height at which some threshold of reflectivity— 20 dBZ for applications in this paper—is observed20 dBZ contour in the interpolated radar dataset. Echo-top heights are determined by finding the highest 20 dBZ echo in a single contiguous cloud echo volume by searching downward from the top of the radar datasetecho domain. We choose a 20 dBZ threshold because high reflectivity, such as 40 dBZ, occurs less frequently and often results in a small sample size of echoes. Because our echo-top detection algorithm starts from the top of the dataset (Sec. 3.3), too small of a threshold may could erroneously classify anvil cloud that extends above a convective element as the top of the convective core. Additionally, a 20 dBZ threshold is easily comparable on the large-scale with data from the TRMM precipitation radar, which has a minimum detectable reflectivity of about 17 dBZ. Lower reflectivity is not detected by TRMM, and radar beams in higher reflectivity regions can be heavily attenuated.

For purposes of discussing the convection during the IOP, we have subjectively classified radar echoes according to their areal coverage of stratiform precipitation. Figure 3 shows typical examples of what we refer herein to as isolated convection, echo clusters, squall lines, and mesoscale convective systems (MCSs). Isolated cells have little associated stratiform, and echo clusters usually have only small areas of associated stratiform. Squall lines are deep convective elements organized into a rapidly propagating convective echo line, which may or may not have an attached stratiform echo. Squall lines are often located along cold pool boundaries that advance laterally from convective downdrafts—in this case with a west-to-east component of motion. MCSs are systems consisting of a combination of widespread stratiform precipitation and deep convective cells, which in some cases are embedded in the stratiform precipitation and in other cases trail behind a squall line. The largest MCSs have stratiform regions hundreds of kilometers in dimension. Extremely large MCSs with such gigantic stratiform regions occurred in TOGA COARE [e.g. Fig. 12 of Chen et al., 1996] as well as in DYNAMO/AMIE, where they were seen both by the S-PolKa radar on Addu Atoll and the shipborne radar on the Revelle (www.atmos.washington.edu/~houze/DYNAMO-AMIE/).

The first LCE began in early October after anomalous westerlies above 500 hPa transporting dry air from eastern Africa and the Arabian Peninsula subsided. Isolated convective cells extending as high as ~8 km were observed through 10 October. By 12 October, individual convective elements began to form into clusters of precipitating echoes, which included convective elements and small, lightly precipitating stratiform areas surrounding them. Figure 2 indicates that, at this time, the amount of stratiform precipitation began to increase. A large MCS moved within range and over S-PolKa on 16 October, and widespread stratiform precipitation was subsequently observed every other day through 30 October. On days during which widespread stratiform echo was not present, the radar observed mostly isolated convective elements and several small clusters of precipitation. A similar two-day periodicity in mesoscale organization was also observed during TOGA COARE and MISMO, during which convection was probably linked to the periodicity of westward propagating inertio-gravity waves [of the type discussed by Takayabu, 1994; Chen et al., 1996; Yamada et al., 2010]. Recent analysis of DYNAMO and AMIE data suggest that the two-day periodicity in precipitation (Fig. 2) at Addu Atoll may also be linked to inertio-gravity waves [Zuluaga and Houze, in press2013]. The two-day timescale may also represent the time required for the atmosphere to again become unstable after a large mesoscale event [Chen and Houze, 1997a,b]. What is new here is that the stratiform component of the precipitation observes follows the two-day variation along with the total rainfall.

The second LCE occurred during November and at first exhibited isolated convective cells and small cloud clusters. A few of these convective elements generated downdrafts strong enough to form cold pools that resulted in squall line formation. After a few isolated convective cells formed on 4-9 November, large clusters began to develop from 10-17 November. Large MCSs occurred on 18, 23, and 27-28 November. The periodicity of widespread stratiform precipitation was notably different during LCE2 than during LCE1. Widespread stratiform precipitation during LCE2 was observed only about every four days, similar to the 3-4 day and 6-8 day periodicity in convection seen over the IO and attributed to mixed-Rossby gravity waves and equatorial Rossby waves during MISMO [Yasunaga et al., 2010]. (The S-band radar on S-PolKa was not operational from 0700 UTC, 20 November-1100 UTC, 21 November. During that time, convection is subjectively characterized using observations from the nearby SMART-R.) On days between individual MCS events, convection mostly took the form of small cloud clusters or squall lines and not isolated convective cells as was the case during LCE1. For the remainder of the paper, "rainy periods" refer to those during a LCE for which the hourly, radar-estimated rainfall averaged over the entire domain was at least 0.1 mm. "Dry periods" are those that fall below the same threshold. A domain averaged hourly rainfall of 0.1 mm is typically found when part of a large stratiform region occupies a small portion along the outer edge of the radar domain or when echo clusters are observed. About two-thirds of the total time during an active LCE is a "rainy period"; thus, the sample size of dry periods and wet periods for a 0.1 mm threshold is high enough so that differences in humidity profiles or convection between the two categories might beare statistically robust. Results in later sections have include some discussion of the sensitivity to the threshold used, and this sensitivity is briefly further explored in Appendix A.

Convection prior to LCE3 mostly consisted of shallow cumulus and some taller isolated convective elements and cloud echo clusters through 7 December. An MCS passed westward over and north of S-PolKa on 8 December; a review of geosynchronous satellite imagery reveals that this MCS was an isolated, local event at the time and was not associated with an MJO LCE. Widespread dry conditions persisted before and after the 8 December MCS. With the exception of the 8 December event, precipitation amounts remained low until 15 December, when several squall lines developed in and/or passed through the radar domain. Some large echo clusters and limited stratiform precipitation were observed during the second half of December. A domain-average of more than 7 mm day^-1 in precipitation was estimated each day between 19 December and 24 December. During this period, squall lines with deep convective cores frequently developed near the radar in proximity to large MCSs that mostly remained outside of the radar domain. As such, the highest precipitation amounts likely occurred outside the range of S-PolKa, and this at least partially explains why less rainfall was observed during LCE3 that during LCE1 and LCE2.